forward-forward-learning

What problem does it solve?

Hinton's Forward-Forward (FF) algorithm enables local, backpropagation-free learning suitable for biologically plausible setups, on-chip training, memory-efficient networks, and parallel layer-wise updates.

Core Features & Use Cases

• Two Forward Passes: Positive and negative data passes to optimize layer-wise goodness.
• Goodness-Based Objectives: Layer-wise objectives using goodness of activations.
• Self-Contrastive Extensions: Self-generated negatives for robust learning.
• On-Chip Learning & Memory Efficiency: Suitable for neuromorphic hardware and edge devices.

Quick Start

Initialize an FFNetwork with your desired layer dimensions and begin training with a local-forward step loop.